Towards Personalizing Shared Autonomy with Human-in-the-Loop Robot Programs

Daphne Chen, Sahitha Vuddagiri, Claire Yang, Nick Walker, Maya Cakmak

Keywords: Shared Autonomy, Human in the Loop, Systems for Teleoperation

TL;DR: We introduce the Shared Autonomy Toolkit, a system that enables a diverse range of users to program robots with human-in-the-loop shared control.

Abstract:

As robots develop increasingly advanced capabilities and become more
broadly used in the home, a key challenge remains in developing systems that ef-
fectively allow people to share control with a robot partner. We introduce the
Shared Autonomy Toolkit, which enables users to specify when and where to as-
sume control in a robot program, enabling them to customize task execution to
their preferences. In a user study with 8 participants across various skill levels, we
investigate how diverse users interact with this system and identify patterns for
when humans choose to insert themselves into the loop in a long-horizon manip-
ulation task. Our findings highlight that integrating users into the loop in shared
autonomy improves task success rates, perceived sense of control, and user ex-
perience, illuminating patterns that can inform how robots learn to adapt across
user preferences. Our framework also serves as a scalable platform for collecting
high-quality, user-driven intervention data, advancing the development of robots
that learn to adapt to humans.

Submission Number: 21